Safety Evaluation of Multiple Strategies at Stop-Controlled Intersections

Chapter 6. Data Collection

SCDOT provided the majority of data for this study. The dataset included the following data elements:

• Treatment implementation data: SCDOT provided information related to treatment sites and start and completion dates for each improvement. SCDOT also provided the research team with work orders, drawings, and sketches for these locations. The research team used some of these additional data to verify the intersection configurations.
• Reference site data: SCDOT provided a list of intersections that had not been treated. The research team used these intersections as potential reference sites.
• Traffic data: SCDOT provided a statewide AADT data file for 2014. This file has information for almost all mainline routes and cross streets. The research team obtained additional AADT files for 2006 to 2014 from SCDOT’s website. These publicly available files do not have all the details necessary (many AADT for minor routes are missing). The research team used these files to calculate traffic growth factors and estimated AADTs for other years.
• Crash data: SCDOT provided the research team with crash data files for 2005 to 2014.

The research team collected additional data using Google® Earth™ and Google® Maps™.

Treatment Sites

SCDOT provided the research team with a list of all intersections under the ground-level contract that includes signing and pavement marking installation (a total of 918 locations.) Because the ground-level contract covers both the improvements at signalized and stop-controlled intersections, the first step was to separate stop-controlled intersections from the file. Key pieces of information from this data file included county, route designations and numbers for both mainline and cross street (e.g., US 25, SC 12), start and completion dates of installation, number of lanes (e.g., two or four lanes on the mainlines, two lanes on the cross street), and area type (i.e., urban, rural). The first step of processing the data was to convert route designation and number for both mainline and cross street into three identification codes as follows:

• Route type code (i.e., US = 2, SC = 4, and S = 7).
• Route number.
• Route auxiliary (i.e., Mainline = 0, Alternate = 2, and Business = 7).

These three identifiers would later be used to link the crash and traffic data files to each intersection. Once crash and traffic data were linked to each intersection, the research team summarized the number of crashes per year by type for each location.

The start and completion dates allowed the team to identify the before and after periods. Before and after periods included complete calendar years during which there was no installation activity. For example, if the work at a given intersection started in December 2009 and was not completed until January 2010, two full calendar years of 2009 and 2010 were considered “installation years” and removed from the dataset. The before period in this research is from 2005 (the first year of available data) to 2008 (the last full year of no construction activity). Similarly, the after period is from 2011 to 2014 (the last year of available data).

The result of this process was a list of stop-controlled intersections with location identifiers in uniform format across different data files. This list included the before and after periods. These intersections were candidates for the treatment group used in the EB evaluation. The research team checked all work orders and work plans to collect number of legs and verify number of lanes for each intersection. The research team conducted a manual process of verifying candidate intersections in Google® Earth™ (i.e., visual verification) to select the final treatment group. The research team flagged intersections—and later dropped treatment sites from the list of candidates—based on the following criteria:

1. The intersection is located close to another intersection or facility (e.g., railroad crossing) and it was not possible to separate crashes and operations, as seen in figure 7. The candidate treatment site is highlighted, however, it is located next to a railroad crossing and what appears to be a major signalized intersection. In this case, the research team determined that it would be difficult to know if a crash occurred because of the intersection of interest or something else. In these cases, the research team dropped the site from the candidate pool.
2. The intersection has an abnormal configuration, and the data fail to reflect the anomaly. This is often the case where an intersection has an extreme skew angle or it is an exit ramp from a limited-access highway. Figure 8 and figure 9 are two examples. Both intersections were coded in the installation data file as three-legged, stop-controlled intersections between the surface street and the limited-access highway above (i.e., US 178 and US 123-Calhoun Memorial Hwy, US 378-Sunset Blvd and SC 12-Jarvis Klapman Blvd). There is no indication in the data file that these are exit ramps. The traffic volumes associated with these locations are for the major limited-access highways, not the ramps themselves. All intersections similar to these were flagged and later dropped from the treatment group.

Following SCDOT’s advice, the research team decided to exclude all intersections in Beaufort County because there were changes in route names and numbers in this county that could lead to inaccurate matching of traffic and crash data.

Imagery ©2016 Google®, Map data ©2016 Google®.
Figure 7. Screenshot. Example 1 of check and verify treatment site in Google® Earth™ (original image modified with circle around intersection).⁽¹⁶⁾

Imagery ©2016 Google®, Map data ©2016 Google®.
Figure 8. Screenshot. Example 2 of check and verify treatment site in Google® Earth™ (original image modified with circle around intersection).⁽¹⁷⁾

Imagery ©2016 Google®, Map data ©2016 Google®.
Figure 9. Screenshot. Example 3 check and verify treatment site in Google® Earth™ (original image modified with circle around intersection).⁽¹⁸⁾

Reference Sites

SCDOT provided the research team with a list of more than 3,000 intersections—both stop-controlled and signalized—for reference sites. Similar to the installation data, this list of intersections included key location identifiers (e.g., county, route designations, and numbers) and intersection characteristics (e.g., number of lanes on the mainline and cross street, area type, and type of traffic control). Therefore, the research team followed similar steps to process the raw data. The route identifiers were converted to a common format to link crash and traffic data for each intersection from different files. However, this data file did not provide number of legs, a key variable for these potential reference sites.

The research team decided to collect number of legs using Google® Earth™ and Google® Maps™. It was infeasible to locate and collect number of legs from Google® Earth™ for every intersection because of resource constraints. Instead, the research team randomly sampled at least 30 intersections for each group from the pool of candidate reference sites, and took the following steps:

1. Separate the pool of stop-controlled intersections into different categories using the available information (e.g., rural intersections with two lanes on the mainline and two lanes on the cross street, urban intersections with four lanes on the mainline).
2. Randomize the order in each intersection category using a random number generator.
3. Start from the top of the list for each category, locate the intersection in Google® Maps™ and Google® Earth™, and determine the number of legs and verify the number of lanes.

The research team repeated these steps until there were at least 30 sites for each group (e.g., three-legged, rural intersections with two mainline lanes and two cross street lanes). Figure 10 shows a screenshot of a four-legged, rural intersection between US 76 (Garners Ferry Rd) and S-69 (Congress Rd) in Richland County, located in Google® Maps™ in satellite image mode.

Imagery ©2016 Google®, Map data ©2016 Google®.
Figure 10. Example of collecting number of legs for reference site from Google® Earth™.⁽¹⁹⁾

Traffic Data

SCDOT provided the research team with a statewide traffic volume data file for 2014. The research team used data in this file and merged with both candidate treatment and reference sites. This data file had more details with AADT information for both mainline and cross street for most intersections. The research team made a request but SCDOT was not able to provide similar data for other years. With SCDOT’s advice, the research team downloaded AADT files for 2006 to 2014 publicly available on SCDOT’s website. However, these data files were much less detailed than the 2014 file the research team received from SCDOT staff. AADT information from these files was not available for many intersection mainlines and a majority of intersection cross streets. The research team used these data files to create growth factors by county. The research team used the growth factors and the detailed 2014 data file to estimate AADT information for 2006 to 2013. AADT for 2005 was not available from SCDOT’s website, so the research team extrapolated 2005 AADT based on data for 2006 to 2008. If AADT for either the mainline or cross street was still missing after the data processing, the research team dropped the intersections from the pools of treatment or reference sites. Note that there were few intersections that fell into this group.

Crash Data

SCDOT provided 10 years of crash data (2005 to 2014). A unique accident number identifies each crash in the data files. A combination of the following variables was used to identify the location of each crash:

• County number (e.g., 1 = Abbeville, 2 = Aiken, and 3 = Allendale).
• Route type code (e.g., 2 = US, 4 = SC, and 7 = S).
• Route number.
• Route auxiliary (e.g., 0 = Mainline, 2 = Alternate, and 7 = Business).
• Crossing route type code (e.g., 2 = US, 4 = SC, and 7 = S).
• Crossing route number.
• Crossing route auxiliary (e.g., 0 = Mainline, 2 = Alternate, and 7 = Business).
• Base distance offset from the intersection (e.g., 1 = 0.01 mi, 5 = 0.05 mi, and 10 = 0.1 mi).

Note that the research team used crossing route in this context as a reference point, and the offset determined the distance from that reference point to the crash location. Route and crossing route in crash data files do not necessarily mean the mainline and minor routes in the same context of an intersection. The route indicates the roadway on which the crash occurred, and the crossing route indicates the crossing street at the nearest intersection (reference point). Both can be the mainline or the minor roads of the intersection used as the reference point.

The research team screened crash location information to identify and count crashes at each intersection. The crash data files did not provide a specific code to determine “intersection-related” crashes. Therefore, the process of locating and counting crashes at each intersection relied solely on crash location. The research team considered a crash “intersection-related” and counted it toward the number of crashes at an intersection if the location information indicated the crash occurred within 0.05 mi (264 ft), as was recommended by SCDOT staff.

The research team used number of fatalities and injuries coded for each crash to determine crash severity. Manner of collision determined rear-end and right-angle crashes. Light condition information was also available and identified nighttime crashes.

Table 13 presents the crash type definitions for South Carolina crash data.

Treatment Cost Data

SCDOT provided actual construction cost data for improvements at more than 800 unsignalized intersections. Intersection construction costs were separated into subtotal pavement marking and signing treatment costs. Each intersection received a package of those treatments appropriate for implementation at the site out of the list of potential treatments. The treatment costs varied at each intersection based on the unique package of treatments it received. Table 14 summarizes the treatment costs.

Note that some intersections only had pavement marking or signing improvements, but all intersections had at least some of one type of treatment installed.

Maintenance costs are dependent on the countermeasures installed at a given intersection. Without a record of the countermeasures installed at each intersection, it is difficult to estimate maintenance costs and service life. In addition, preliminary engineering (PE) costs were not supplied by SCDOT. For systemic projects, PE costs often represent 10 to 30 percent of the total project costs.

Data Characteristics and Summary

Table 15 and table 16 provide summary information for the data collected for the treatment and reference sites. The information in table 15 should not be used to make simple before–after comparisons of crashes per site-year since it does not account for factors, other than the strategy, that may cause a change in safety between the before and after periods. Such comparisons are properly done with the EB analysis as presented later.